Как создать микшер нестандартной модели в Lightwood

Введение

Микшеры — это центральная часть лайтвуда, задача которой — изучение сопоставления между закодированной функцией и целевым представлением. Lightwood — это среда AutoML, позволяющая создавать и настраивать конвейеры машинного обучения с помощью декларативного синтаксиса под названием JSON-AI. Микшеры учатся отображать закодированное представление, они являются ядром AutoML Lightwood. В этом уроке мы попытаемся реализовать LightGBM как микшер, который обрабатывает категориальные и целочисленные цели.

Установка и настройка

Вы можете установить Lightwood следующим образом:

pip install lightwood

Настройка среды VSCode

Установите и включите настройку синхронизации с помощью учетной записи GitHub (если вы используете несколько компьютеров).
Установите pylance (для типов) и обязательно отключите пирайт
Перейдите к Python > Lint: Enabled и отключите все, кроме flake8.
Установите python.linting.flake8Path на полный путь к flake8 (который flake8)
Установите Python › Formatting: Provider на autopep8
Добавьте -global-config=<path_to>/lightwood/.flake8 и -experimental к Python › Formatting: Autopep8 Args
Установите интерактивный обмен и интерактивную доску для интерактивного обмена

Примечание. Запускайте в виртуальной среде Python с версией Python 3.8 или 3.9.

Шаг 1: Интерфейс микшера

Интерфейс микшера определяется классом BaseMixer, микшеру нужны методы для 4 задач: * подгонка (fit) * прогнозирование (__call__) * построение (__init__) * частичная подгонка (partial_fit), хотя это необязательно

Шаг 2: Пишем наш микшер

Я собираюсь создать файл с именем lightgbm_mixer.py в нашей виртуальной среде. Внутри него мы будем реализовывать наш микшер со следующими строками кода.

from lightwood.mixer import BaseMixer
from lightwood.api.types import PredictionArguments
from lightwood.data.encoded_ds import EncodedDs, ConcatedEncodedDs
from lightwood import dtype
from lightwood.encoder import BaseEncoder
import torch
import pandas as pd
import lightgbm
class LightGBM(BaseMixer):
    model: lightgbm.LGBMClassifier
    def __init__(self, stop_after: int, dtype_dict: dict, target: str, target_encoder: BaseEncoder):
        super().__init__(stop_after)
        self.target_encoder = target_encoder
# Throw in case someone tries to use this for a problem that's not classification, I'd fail anyway, but this way the error message is more intuitive
        if dtype_dict[target] not in (dtype.categorical, dtype.integer):
            raise Exception(f'This mixer can only be used for classification problems! Got target dtype {dtype_dict[target]} instead!')
            # We could also initialize this in `fit` if some of the parameters depend on the input data, since `fit` is called exactly once
        self.model = lightgbm.LGBMClassifier(max_depth=30)
        def fit(self, train_data: EncodedDs, dev_data: EncodedDs) -> None:
            X, Y = [], []
        # By default mixers get some train data and a bit of dev data on which to do early stopping or hyper parameter optimization. For this mixer, we don't need dev data, so we're going to concat the two in order to get more training data. Then, we're going to turn them into an sklearn friendly foramat.
            for x, y in ConcatedEncodedDs([train_data, dev_data]):
                X.append(x.tolist())
                Y.append(y.tolist())
            self.model.fit(X, Y)
    def __call__(self, ds: EncodedDs,
                 args: PredictionArguments = PredictionArguments()) -> pd.DataFrame:
        # Turn the data into an sklearn friendly format
        X = []
        for x, _ in ds:
            X.append(x.tolist())
        Yh = self.model.predict(X)
        # Lightwood encoders are meant to decode torch tensors, so we have to cast the predictions first
        decoded_predictions = self.target_encoder.decode(torch.Tensor(Yh))
        # Finally, turn the decoded predictions into a dataframe with a single column called `prediction`. This is the standard behaviour all lightwood mixers use
        ydf = pd.DataFrame({'prediction': decoded_predictions})
        return ydf
    # We'll skip implementing `partial_fit`, thus making this mixer unsuitable for online training tasks

Шаг 3: Использование нашего микшера

Мы собираемся использовать наш микшер, чтобы определить силу нашей покерной комбинации на основе мастей и рангов всех пяти карт:

https://raw.githubusercontent.com/Tes-program/lightwood_tutorial/main/poker-hand-training.csv

Во-первых, поскольку мы не хотим писать Json AI для этого набора данных с нуля, мы позволим lightwood сгенерировать его автоматически.

from lightwood.api.high_level import ProblemDefinition, json_ai_from_problem, load_custom_module
import pandas as pd
import lightgbm_mixer
# load the code
load_custom_module('lightgbm_mixer.py')
# read dataset
df = pd.read_csv('<https://raw.githubusercontent.com/Tes-program/lightwood_tutorial/main/poker-hand-training.csv>')
# define the predictive task
pdef = ProblemDefinition.from_dict({
    'target': 'Poker_Hand', # column you want to predict
})
# generate the Json AI intermediate representation from the data and its corresponding settings
json_ai = json_ai_from_problem(df, problem_definition=pdef)
# Print it (you can also put it in a file and edit it there)
print(json_ai.to_json())

Ожидаемый результат:

╰─ python predictor.py                                                       ─╯
INFO:lightwood-17958:No torchvision detected, image helpers not supported.
INFO:lightwood-17958:No torchvision/pillow detected, image encoder not supported
/home/teslim/test/lib/python3.8/site-packages/gluonts/json.py:101: UserWarning: Using `json`-module for json-handling. Consider installing one of `orjson`, `ujson` to speed up serialization and deserialization.
  warnings.warn(
/home/teslim/test/lib/python3.8/site-packages/gluonts/model/deepar/__init__.py:18: FutureWarning: The module gluonts.model.deepar has been moved to gluonts.mx.model.deepar. In GluonTS v0.12 it will be no longer possible to use the old path. Try to use 'from gluonts.mx import DeepAREstimator'.
  warnings.warn(
INFO:lightwood-17958:Analyzing a sample of 11020
INFO:lightwood-17958:from a total population of 25010, this is equivalent to 44.1% of your data.
INFO:lightwood-17958:Using 3 processes to deduct types.
INFO:lightwood-17958:Infering type for: Suit_of_Card_1
INFO:lightwood-17958:Infering type for: Rank_of_Card_1
INFO:lightwood-17958:Infering type for: Suit_of_Card_2
INFO:lightwood-17958:Column Suit_of_Card_2 has data type categorical
INFO:lightwood-17958:Infering type for: Rank_of_Card_2
INFO:lightwood-17958:Column Suit_of_Card_1 has data type categorical
INFO:lightwood-17958:Column Rank_of_Card_1 has data type integer
INFO:lightwood-17958:Infering type for: Suit_of_Card_3
INFO:lightwood-17958:Infering type for: Rank_of_Card_3
INFO:lightwood-17958:Column Suit_of_Card_3 has data type categorical
INFO:lightwood-17958:Infering type for: Suit_of_Card_4
INFO:lightwood-17958:Column Rank_of_Card_2 has data type integer
INFO:lightwood-17958:Infering type for: Rank_of_Card_4
INFO:lightwood-17958:Column Rank_of_Card_3 has data type integer
INFO:lightwood-17958:Infering type for: Suit_of_Card_5
INFO:lightwood-17958:Column Suit_of_Card_4 has data type categorical
INFO:lightwood-17958:Infering type for: Rank_of_Card_5
INFO:lightwood-17958:Column Suit_of_Card_5 has data type categorical
INFO:lightwood-17958:Infering type for: Poker_Hand
INFO:lightwood-17958:Column Rank_of_Card_4 has data type integer
INFO:lightwood-17958:Column Rank_of_Card_5 has data type integer
INFO:lightwood-17958:Column Poker_Hand has data type integer
INFO:lightwood-17958:Starting statistical analysis
INFO:lightwood-17958:Finished statistical analysis
{
    "encoders": {
        "Poker_Hand": {
            "module": "NumericEncoder",
            "args": {
                "is_target": "True",
                "positive_domain": "$statistical_analysis.positive_domain"
            }
        },
        "Suit_of_Card_1": {
            "module": "OneHotEncoder",
            "args": {}
        },
        "Rank_of_Card_1": {
            "module": "NumericEncoder",
            "args": {}
        },
        "Suit_of_Card_2": {
            "module": "OneHotEncoder",
            "args": {}
        },
        "Rank_of_Card_2": {
            "module": "NumericEncoder",
            "args": {}
        },
        "Suit_of_Card_3": {
            "module": "OneHotEncoder",
            "args": {}
        },
        "Rank_of_Card_3": {
            "module": "NumericEncoder",
            "args": {}
        },
        "Suit_of_Card_4": {
            "module": "OneHotEncoder",
            "args": {}
        },
        "Rank_of_Card_4": {
            "module": "NumericEncoder",
            "args": {}
        },
        "Suit_of_Card_5": {
            "module": "OneHotEncoder",
            "args": {}
        },
        "Rank_of_Card_5": {
            "module": "NumericEncoder",
            "args": {}
        }
    },
    "dtype_dict": {
        "Suit_of_Card_1": "categorical",
        "Rank_of_Card_1": "integer",
        "Suit_of_Card_2": "categorical",
        "Rank_of_Card_2": "integer",
        "Suit_of_Card_3": "categorical",
        "Rank_of_Card_3": "integer",
        "Suit_of_Card_4": "categorical",
        "Rank_of_Card_4": "integer",
        "Suit_of_Card_5": "categorical",
        "Rank_of_Card_5": "integer",
        "Poker_Hand": "integer"
    },
    "dependency_dict": {},
    "model": {
        "module": "BestOf",
        "args": {
            "submodels": [
                {
                    "module": "Neural",
                    "args": {
                        "fit_on_dev": true,
                        "stop_after": "$problem_definition.seconds_per_mixer",
                        "search_hyperparameters": true
                    }
                },
                {
                    "module": "LightGBM",
                    "args": {
                        "stop_after": "$problem_definition.seconds_per_mixer",
                        "fit_on_dev": true
                    }
                },
                {
                    "module": "Regression",
                    "args": {
                        "stop_after": "$problem_definition.seconds_per_mixer"
                    }
                },
                {
                    "module": "RandomForest",
                    "args": {
                        "stop_after": "$problem_definition.seconds_per_mixer",
                        "fit_on_dev": true
                    }
                }
            ]
        }
    },
    "problem_definition": {
        "target": "Poker_Hand",
        "pct_invalid": 2,
        "unbias_target": true,
        "seconds_per_mixer": 42768.0,
        "seconds_per_encoder": null,
        "expected_additional_time": 2.305413007736206,
        "time_aim": 259200,
        "target_weights": null,
        "positive_domain": false,
        "timeseries_settings": {
            "is_timeseries": false,
            "order_by": null,
            "window": null,
            "group_by": null,
            "use_previous_target": true,
            "horizon": null,
            "historical_columns": null,
            "target_type": "",
            "allow_incomplete_history": true,
            "eval_incomplete": false,
            "interval_periods": []
        },
        "anomaly_detection": false,
        "use_default_analysis": true,
        "ignore_features": [],
        "fit_on_all": true,
        "strict_mode": true,
        "seed_nr": 1
    },
    "identifiers": {},
    "imputers": [],
    "accuracy_functions": [
        "r2_score"
    ]
}

Теперь нам нужно отредактировать ключ mixer's этого JSON ai, чтобы указать лайтвуду использовать наш пользовательский микшер. Мы можем использовать его вместе с другими, а также объединить с ними в конце или отдельно. Но в этом случае я заменю все существующие микшеры синтаксисом ниже

json_ai.model['submodels'] = [{
    'module': 'lightgbm',
    
}]

Затем мы сгенерируем некоторый код и, наконец, превратим этот код в объект-предиктор и подгоним его к исходным данным со следующим синтаксисом:

from lightwood.api.high_level import code_from_json_ai, predictor_from_code
code = code_from_json_ai(json_ai)
predictor = predictor_from_code(code)

Ожидаемый результат должен быть:

Теперь давайте обучим нашу модель предсказаниям с помощью:

predictor.learn(df)

Ваш ожидаемый результат должен быть:

INFO:lightwood-69350:No torchvision detected, image helpers not supported.
INFO:lightwood-69350:No torchvision/pillow detected, image encoder not supported
INFO:lightwood-69350:Analyzing a sample of 11020
INFO:lightwood-69350:from a total population of 25010, this is equivalent to 44.1% of your data.
INFO:lightwood-69350:Using 3 processes to deduct types.
INFO:lightwood-69350:Infering type for: Suit_of_Card_1
INFO:lightwood-69350:Infering type for: Rank_of_Card_1
INFO:lightwood-69350:Infering type for: Suit_of_Card_2
INFO:lightwood-69350:Column Suit_of_Card_1 has data type categorical
INFO:lightwood-69350:Infering type for: Rank_of_Card_2
INFO:lightwood-69350:Column Suit_of_Card_2 has data type categorical
INFO:lightwood-69350:Column Rank_of_Card_1 has data type integer
INFO:lightwood-69350:Infering type for: Suit_of_Card_3
INFO:lightwood-69350:Infering type for: Rank_of_Card_3
INFO:lightwood-69350:Column Suit_of_Card_3 has data type categorical
INFO:lightwood-69350:Infering type for: Suit_of_Card_4
INFO:lightwood-69350:Column Rank_of_Card_2 has data type integer
INFO:lightwood-69350:Infering type for: Rank_of_Card_4
INFO:lightwood-69350:Column Rank_of_Card_3 has data type integer
INFO:lightwood-69350:Infering type for: Suit_of_Card_5
INFO:lightwood-69350:Column Suit_of_Card_4 has data type categorical
INFO:lightwood-69350:Infering type for: Rank_of_Card_5
INFO:lightwood-69350:Column Rank_of_Card_4 has data type integer
INFO:lightwood-69350:Infering type for: Poker_Hand
INFO:lightwood-69350:Column Rank_of_Card_5 has data type integer
INFO:lightwood-69350:Column Suit_of_Card_5 has data type categorical
INFO:lightwood-69350:Column Poker_Hand has data type integer
INFO:lightwood-69350:Starting statistical analysis
INFO:lightwood-69350:Finished statistical analysis
INFO:lightwood-69350:Unable to import black formatter, predictor code might be a bit ugly.
INFO:lightwood-69350:[Learn phase 1/8] - Statistical analysis
INFO:lightwood-69350:Starting statistical analysis
INFO:lightwood-69350:Finished statistical analysis
DEBUG:lightwood-69350: `analyze_data` runtime: 1.3 seconds
INFO:lightwood-69350:[Learn phase 2/8] - Data preprocessing
INFO:lightwood-69350:Cleaning the data
DEBUG:lightwood-69350: `preprocess` runtime: 0.51 seconds
INFO:lightwood-69350:[Learn phase 3/8] - Data splitting
INFO:lightwood-69350:Splitting the data into train/test
DEBUG:lightwood-69350: `split` runtime: 0.07 seconds
INFO:lightwood-69350:[Learn phase 4/8] - Preparing encoders
INFO:lightwood-69350:Encoding UNKNOWN categories as index 0
INFO:lightwood-69350:Encoding UNKNOWN categories as index 0
INFO:lightwood-69350:Done running for: Suit_of_Card_1
INFO:lightwood-69350:Encoding UNKNOWN categories as index 0
INFO:lightwood-69350:Done running for: Rank_of_Card_1
INFO:lightwood-69350:Done running for: Suit_of_Card_2
INFO:lightwood-69350:Encoding UNKNOWN categories as index 0
INFO:lightwood-69350:Done running for: Rank_of_Card_2
INFO:lightwood-69350:Done running for: Suit_of_Card_3
INFO:lightwood-69350:Done running for: Rank_of_Card_3
INFO:lightwood-69350:Done running for: Suit_of_Card_4
INFO:lightwood-69350:Encoding UNKNOWN categories as index 0
INFO:lightwood-69350:Done running for: Rank_of_Card_4
INFO:lightwood-69350:Done running for: Suit_of_Card_5
INFO:lightwood-69350:Done running for: Rank_of_Card_5
DEBUG:lightwood-69350: `prepare` runtime: 0.76 seconds
INFO:lightwood-69350:[Learn phase 5/8] - Feature generation
INFO:lightwood-69350:Featurizing the data
DEBUG:lightwood-69350: `featurize` runtime: 0.0 seconds
INFO:lightwood-69350:[Learn phase 6/8] - Mixer training
INFO:lightwood-69350:Training the mixers
INFO:lightwood-69350:Loss of 66.96073913574219 with learning rate 0.0001
INFO:lightwood-69350:Loss of 73.57966613769531 with learning rate 0.00014
INFO:lightwood-69350:Found learning rate of: 0.0001
INFO:lightwood-69350:Loss @ epoch 1: 69.4616584777832
INFO:lightwood-69350:Loss @ epoch 2: 65.4459950373723
INFO:lightwood-69350:Loss @ epoch 3: 60.80090155968299
INFO:lightwood-69350:Loss @ epoch 4: 55.934378990760216
INFO:lightwood-69350:Loss @ epoch 5: 51.119818467360275
INFO:lightwood-69350:Loss @ epoch 6: 46.61281350942758
INFO:lightwood-69350:Loss @ epoch 7: 42.459257272573616
INFO:lightwood-69350:Loss @ epoch 8: 39.36399899996244
INFO:lightwood-69350:Loss @ epoch 9: 37.174588423508865
INFO:lightwood-69350:Loss @ epoch 10: 35.81990872896635
INFO:lightwood-69350:Loss @ epoch 11: 35.092872913067154
INFO:lightwood-69350:Loss @ epoch 12: 34.748955946702225
INFO:lightwood-69350:Loss @ epoch 13: 34.594378838172325
INFO:lightwood-69350:Loss @ epoch 14: 34.534512739915115
INFO:lightwood-69350:Loss @ epoch 15: 34.50775733360877
INFO:lightwood-69350:Loss @ epoch 16: 34.49432549109826
INFO:lightwood-69350:Loss @ epoch 17: 34.48627471923828
INFO:lightwood-69350:Loss @ epoch 18: 34.47985076904297
INFO:lightwood-69350:Loss @ epoch 19: 34.473904829758865
INFO:lightwood-69350:Loss @ epoch 20: 34.46845861581656
INFO:lightwood-69350:Loss @ epoch 21: 34.46300271841196
INFO:lightwood-69350:Loss @ epoch 22: 34.45746994018555
INFO:lightwood-69350:Loss @ epoch 23: 34.45192072941707
INFO:lightwood-69350:Loss @ epoch 24: 34.445784055269684
INFO:lightwood-69350:Loss @ epoch 25: 34.43961539635291
INFO:lightwood-69350:Loss @ epoch 26: 34.43367033738356
INFO:lightwood-69350:Loss @ epoch 27: 34.427734375
INFO:lightwood-69350:Loss @ epoch 28: 34.421786088209885
INFO:lightwood-69350:Loss @ epoch 29: 34.41596603393555
INFO:lightwood-69350:Loss @ epoch 30: 34.409483396089996
INFO:lightwood-69350:Loss @ epoch 31: 34.403370490440956
INFO:lightwood-69350:Loss @ epoch 32: 34.39738435011644
INFO:lightwood-69350:Loss @ epoch 33: 34.39150678194486
INFO:lightwood-69350:Loss @ epoch 34: 34.3857788672814
INFO:lightwood-69350:Loss @ epoch 35: 34.380235818716194
INFO:lightwood-69350:Loss @ epoch 36: 34.3740595304049
INFO:lightwood-69350:Loss @ epoch 37: 34.36870927077074
INFO:lightwood-69350:Loss @ epoch 38: 34.36332027728741
INFO:lightwood-69350:Loss @ epoch 39: 34.358152829683746
INFO:lightwood-69350:Loss @ epoch 40: 34.3531012901893
INFO:lightwood-69350:Loss @ epoch 41: 34.34839307344877
INFO:lightwood-69350:Loss @ epoch 42: 34.342967106745796
INFO:lightwood-69350:Loss @ epoch 43: 34.338902106651894
INFO:lightwood-69350:Loss @ epoch 44: 34.33445358276367
INFO:lightwood-69350:Loss @ epoch 45: 34.33023922259991
INFO:lightwood-69350:Loss @ epoch 46: 34.326259026160606
INFO:lightwood-69350:Loss @ epoch 47: 34.32264416034405
INFO:lightwood-69350:Loss @ epoch 48: 34.31824199969952
INFO:lightwood-69350:Loss @ epoch 49: 34.31558961134691
INFO:lightwood-69350:Loss @ epoch 50: 34.312175457294174
INFO:lightwood-69350:Loss @ epoch 51: 34.30907880342924
INFO:lightwood-69350:Loss @ epoch 52: 34.30618432851938
INFO:lightwood-69350:Loss @ epoch 53: 34.30370506873498
INFO:lightwood-69350:Loss @ epoch 54: 34.30031438974234
INFO:lightwood-69350:Loss @ epoch 55: 34.29912596482497
INFO:lightwood-69350:Loss @ epoch 56: 34.29673913808969
INFO:lightwood-69350:Loss @ epoch 57: 34.294622274545524
INFO:lightwood-69350:Loss @ epoch 58: 34.29275483351488
INFO:lightwood-69350:Loss @ epoch 59: 34.291160583496094
INFO:lightwood-69350:Loss @ epoch 60: 34.288580674391525
INFO:lightwood-69350:Loss @ epoch 61: 34.288435422457184
INFO:lightwood-69350:Loss @ epoch 62: 34.28676135723408
INFO:lightwood-69350:Loss @ epoch 63: 34.28536576491136
INFO:lightwood-69350:Loss @ epoch 64: 34.2841430077186
INFO:lightwood-69350:Loss @ epoch 65: 34.283219850980316
INFO:lightwood-69350:Loss @ epoch 66: 34.28125528188852
INFO:lightwood-69350:Loss @ epoch 67: 34.281844505896935
INFO:lightwood-69350:Loss @ epoch 68: 34.280723278339096
INFO:lightwood-69350:Loss @ epoch 69: 34.279798654409554
INFO:lightwood-69350:Loss @ epoch 70: 34.27906007033128
INFO:lightwood-69350:Loss @ epoch 71: 34.27853628305289
INFO:lightwood-69350:Loss @ epoch 72: 34.27686661940355
INFO:lightwood-69350:Loss @ epoch 73: 34.27788866483248
INFO:lightwood-69350:Loss @ epoch 74: 34.276988102839546
INFO:lightwood-69350:Loss @ epoch 75: 34.27630556546725
INFO:lightwood-69350:Loss @ epoch 76: 34.27573101337139
INFO:lightwood-69350:Loss @ epoch 77: 34.275352478027344
INFO:lightwood-69350:Loss @ epoch 78: 34.27378463745117
INFO:lightwood-69350:Loss @ epoch 79: 34.27494137103741
INFO:lightwood-69350:Loss @ epoch 1: 34.26769755146291
INFO:lightwood-69350:Loss @ epoch 2: 34.247992751621965
INFO:lightwood-69350:Loss @ epoch 3: 34.248576589150005
INFO:lightwood-69350:Loss @ epoch 4: 34.24419905407594
INFO:lightwood-69350:Loss @ epoch 5: 34.25551952701984
DEBUG:lightwood-69350: `fit_mixer` runtime: 355.49 seconds
INFO:lightwood-69350:Started fitting LGBM model
INFO:lightwood-69350:A single GBM iteration takes 0.14085173606872559 seconds
INFO:lightwood-69350:Training GBM (<module 'optuna.integration.lightgbm' from '/home/teslim/test/lib/python3.8/site-packages/optuna/integration/lightgbm.py'>) with 121450 iterations given 42766.44290494919 seconds constraint
INFO:lightwood-69350:Lightgbm model contains 822 weak estimators
INFO:lightwood-69350:Updating lightgbm model with 51.0 iterations
INFO:lightwood-69350:Model now has a total of 823 weak estimators
DEBUG:lightwood-69350: `fit_mixer` runtime: 134.86 seconds
INFO:lightwood-69350:Fitting Linear Regression model
INFO:lightwood-69350:Regression based correlation of: 0.3340088172833158
DEBUG:lightwood-69350: `fit_mixer` runtime: 0.79 seconds
INFO:lightwood-69350:Started fitting RandomForest model
INFO:lightwood-69350:The number of trials (Optuna) is 20.
INFO:lightwood-69350:RandomForest parameters of the best trial: {'num_estimators': 482, 'max_depth': 15, 'min_samples_split': 11, 'min_samples_leaf': 1, 'max_features': 0.7093681838916794}
INFO:lightwood-69350:RandomForest based correlation of (train data): 0.3944169749039024
INFO:lightwood-69350:RandomForest based correlation of (dev data): 0.3880994368278863
DEBUG:lightwood-69350: `fit_mixer` runtime: 330.04 seconds
INFO:lightwood-69350:Ensembling the mixer
INFO:lightwood-69350:Mixer: Neural got accuracy: 0.0
INFO:lightwood-69350:Mixer: LightGBM got accuracy: 0.55
INFO:lightwood-69350:Mixer: Regression got accuracy: 0.0
INFO:lightwood-69350:Mixer: RandomForest got accuracy: 0.0
INFO:lightwood-69350:Picked best mixer: LightGBM
DEBUG:lightwood-69350: `fit` runtime: 846.97 seconds
INFO:lightwood-69350:[Learn phase 7/8] - Ensemble analysis
INFO:lightwood-69350:Analyzing the ensemble of mixers
INFO:lightwood-69350:The block ICP is now running its analyze() method
INFO:lightwood-69350:The block ConfStats is now running its analyze() method
INFO:lightwood-69350:The block AccStats is now running its analyze() method
INFO:lightwood-69350:The block PermutationFeatureImportance is now running its analyze() method
INFO:lightwood-69350:[PFI] Using a random sample (1000 rows out of 2501).
INFO:lightwood-69350:[PFI] Set to consider first 10 columns out of 10: ['Suit_of_Card_1', 'Rank_of_Card_1', 'Suit_of_Card_2', 'Rank_of_Card_2', 'Suit_of_Card_3', 'Rank_of_Card_3', 'Suit_of_Card_4', 'Rank_of_Card_4', 'Suit_of_Card_5', 'Rank_of_Card_5'].
DEBUG:lightwood-69350: `analyze_ensemble` runtime: 4.15 seconds
INFO:lightwood-69350:[Learn phase 8/8] - Adjustment on validation requested
INFO:lightwood-69350:Updating the mixers
torch.cuda.amp.GradScaler is enabled, but CUDA is not available.  Disabling.
INFO:lightwood-69350:Loss @ epoch 1: 34.25837150506214
INFO:lightwood-69350:Loss @ epoch 2: 34.26038431488307
INFO:lightwood-69350:Loss @ epoch 3: 34.259460753044195
INFO:lightwood-69350:Loss @ epoch 4: 34.26046121647928
INFO:lightwood-69350:Loss @ epoch 5: 34.260302923421946
INFO:lightwood-69350:Model now has a total of 824 weak estimators
INFO:lightwood-69350:Fitting Linear Regression model
INFO:lightwood-69350:Regression based correlation of: 0.3338930435705283
DEBUG:lightwood-69350: `adjust` runtime: 28.23 seconds
DEBUG:lightwood-69350: `learn` runtime: 882.1 seconds

Наконец, мы можем использовать обученный предиктор, чтобы сделать некоторые прогнозы:

predictions = predictor.predict(pd.DataFrame({
    'Suit_of_Card_1': [1, 2, 3],
    'Rank_of_Card_1': [10, 11, 12],
    'Suit_of_Card_2': [1, 2, 3],
    'Rank_of_Card_2': [11, 13, 11],
    'Suit_of_Card_3': [1, 2, 3],
    'Rank_of_Card_3': [1, 2, 3],
    'Suit_of_Card_4': [1, 2, 3],
    'Rank_of_Card_4': [12, 12, 10],
		'Suit_of_Card_5': [1, 2, 3],
		'Rank_of_Card_5': [1, 1, 1]
}))
print(predictions)

Вот и все, что нужно для решения задачи прогнозирования с использованием светлой древесины с помощью собственного микшера.

Что дальше?

Получайте удовольствие, пробуя это сами!

Пометьте репозиторий MindsDB на GitHub.
Зарегистрируйтесь для получения бесплатной учетной записи MindsDB
Присоединяйтесь к сообществу MindsDB в Slack или GitHub, чтобы задавать вопросы и делиться своими идеями и мыслями.

Поставьте лайк или оставьте комментарий, если этот урок был полезен